Field report S-19 / P5 · sheet /strategy
Corporate StrategyFiled from the territory
Alibaba Launches Qwen4-Max, Claims Parity With OpenAI and Anthropic
Alibaba's Qwen4-Max arrives with 2.4 trillion parameters, a one-million-token context window and a 62% cut in average task costs, claiming parity with OpenAI and Anthropic.
- Grid
- S-19 / P5
- Surveyed
- November 20, 2026
- Reading
- 3 min read
- Correspondent
- Amara Osei

Field notes
- Alibaba launched Qwen4-Max on Monday, a 2.4-trillion-parameter model with a one-million-token context window.
- Average cost per task on Alibaba's Work360 benchmark fell 62%, from $4.43 to $1.67.
- Qwen4-Max scored 52.69 on Work360, up from 47.74 for Qwen3.8-Max.
- The share of securely generated code rose from 8% to 14.7% after new security checks in training.
- Qwen4-Max ships without open weights, though smaller Qwen4 models will be open-sourced.
Alibaba released Qwen4-Max on Monday, a 2.4-trillion-parameter flagship AI model that the company says matches systems from OpenAI and Anthropic, according to Bloomberg. The launch sharpens the competition between Chinese and American AI developers at the top of the model market.
The release pairs an aggressive capability claim with an equally aggressive price cut. On Alibaba's Work360 benchmark for industry tasks, Qwen4-Max scored 52.69, up from 47.74 for its predecessor, Qwen3.8-Max. The average cost per task in that evaluation fell from $4.43 to $1.67 — a 62% reduction.
What does the new model claim to do?
Alibaba said Qwen4-Max placed close to the best results on coding tests and beat Anthropic's Claude Fable 5.1 and Moonshot's Kimi K3.1 Pro. The comparison to Anthropic and OpenAI systems comes via Bloomberg's reporting on the company's assertions.
The model runs on a new Qwen4 architecture with three headline specifications:
- 2.4 trillion parameters (the count of trainable values inside the model)
- native vision capabilities
- a one-million-token context window, which determines how much text and data the model can process at once
Alibaba said it lowered API prices compared with Qwen3.8-Max, with the largest cuts applied to cached-input pricing, where previously processed input data is reused at a discount.
Where can users access it?
Availability spans both consumer-facing and developer channels. Users can reach the model through QwenCloud with either a Token Plan or an API (an application programming interface, the direct connection developers use to build the model into their own software). Enterprise access runs through QwenWork, Qoder and QwenStudio.
How does Alibaba handle code security?
The company said it added code security checks to the model's software engineering training. The share of code generated securely at the function, file and repository levels rose from 8% to 14.7%, according to Alibaba.
For cybersecurity, Alibaba drew a defensive line in its training data. It kept data for defensive tasks such as vulnerability detection and remediation, and removed data for offensive tasks. An internal expert review found no capabilities posing large-scale cybersecurity risks, the company said.
Why does the closed release matter?
Alibaba released Qwen4-Max without open weights, according to Bloomberg. Weights are the trained numerical values that define a model's behavior; releasing them would let outside users download the model or tailor it to their own needs. The closed approach for the flagship contrasts with the strategy that built the Qwen series' reputation.
The Qwen family has been downloaded more often than any other open-weight model family worldwide, according to Bloomberg. Alibaba said smaller models in the Qwen4 family will come to open source, preserving that pipeline while keeping its top-tier system proprietary.
Who else is competing from China?
Moonshot and DeepSeek are among the other Chinese firms releasing models that have quickly grown stronger at programming and logical problem-solving, Bloomberg reported. Moonshot's Kimi K3.1 Pro already appears in Alibaba's own benchmark comparisons, underscoring how directly the domestic rivals measure themselves against one another.
The strategic logic behind the launch is visible in the numbers: a 62% cost reduction per task, a 52.69 benchmark score and a 2.4-trillion-parameter architecture combine capability, scale and price pressure in a single release. The closed weights on the flagship signal that Alibaba intends to compete at the frontier on commercial terms rather than on openness alone.
via qz.com (Source)
More from Amara Osei
Show full bio
Market editor covering people, launches and day-to-day practice in corporate strategy and chief-executive leadership at CEO Frontier.
479 articles
Adjacent sheets
- Samsung Tests $4.99 Monthly Max Plan for Galaxy AI Features
- Chinese AI Models Now Six to 12 Months Behind U.S., Says Kai-Fu Lee
- TSMC Posts 59.7% October Revenue Jump as AI Spending Accelerates
- Anthropic IPO Asks Investors to Bet on AI That Builds the Company Itself
- Claude Fable 5.2 and Mythos 5.2 undercut GPT-6 Astra by up to 80%